Classification of mental tasks using de-noised EEG signals

The wavelet based de-noising can be employed with the combination of different kind of threshold parameters, threshold operators, mother wavelets and threshold rescaling methods. The central issue in wavelet based de-noising method is the selection of an appropriate threshold parameters. If the threshold is too small, the signal is still noisy but if it is too large, important signal features might lost. This study will investigate the effectiveness of four types of threshold parameters i.e. threshold selections based on Stein´s unbiased risk estimate (SURE). Universal, heuristic and minimax, autoregressive Burg model with order six is employed to extract relevant features from the clean signals. These features are classified into five classes of mental tasks via an artificial neural network. The results show that the rate of correct classification varies with different thresholds. From this study, it shows that the de-noised EEG signal with heuristic threshold selection outperforms the others. Soft thresholding procedure and sym8 as the mother wavelet are adopted in this study.